The Agent Skills Directory

[COMMAND_EXECUTION]: Orchestrates various Model Context Protocol (MCP) tools for academic searching and document parsing, as well as a runSubagent tool for delegated analysis tasks.\n- [DATA_EXFILTRATION]: Accesses external academic APIs (PubMed) and local file systems to retrieve and index research materials as part of its core functionality.\n- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection when processing external literature.\n
Ingestion points: External PDFs and full-text articles ingested through tools like ingest_documents and get_fulltext.\n
Boundary markers: The workflow does not explicitly specify delimiters or isolation markers for content passed to analysis agents.\n
Capability inventory: The skill has permissions to write analysis files and execute subagents.\n
Sanitization: There is no evidence of sanitization for ingested text, although the use of subagents for analysis provides architectural isolation.

literature-review